x-octo home Business judgment on AI products
中文

Business judgment on AI products

AI 搜索|百度

Keep watching

Turns Baidu search into a chat door: it answers, digs further, and keeps writing and image tools on the same page.

Category is set Has usage data General assistantsMonthly visits 44.85MMoM +14%
First tracked here
2026-08-11
Last updated here
2026-08-11

01

Why this would be needed

Start inside the user's day · Public facts + observable behavior · 2026-09-23

Use case

On desktop, an ordinary user facing a multi-step question or a content task hands it to Baidu AI Search and, on one page, follows up, reads sources, writes or generates images, ending with a usable answer or asset.

Traditional Baidu keyword search with manual filtering, plus separate writing or image tools; some users switch to other conversational AI assistants.

Legacy keyword search returns a list of links, so the user must open, filter and stitch sources manually and then switch to another tool to write or make images; a multi-step task is scattered across tabs, with duplicated work and lost context as the main friction.

xOcto's call

A case study in how an incumbent does AI — the traffic is real, and so is the anxiety.

The trend is the search box becoming a conversation. The entry is not generic Q&A — start with office lookup, homework help, and local life, where Baidu already has users. Free answers fighting ad revenue is the structural risk.

Reason to use it

Why users would choose it

Inference: versus opening links one by one and switching tools, it chains follow-up, reading and creation on one page, removing the tab-switching and context re-describing steps, so desktop work and study users who research and write in the same session would pick it for complex questions; public material offers no retention or payment evidence.

Where the easy answer breaks down

The tension worth following

① Where the 12M subscriber count goes — willingness to pay is the hardest proof metric; for AI search; ② Whether traffic holds through the "Wenxin Assistant" rebrand — the transition will; expose whether the product has its own mindshare; ③ Whether the RMB 25M/day agent revenue keeps growing — it de…

If this is your job

Worth dissecting. Inference: versus opening links one by one and switching tools, it chains follow-up, reading and creation on one page, removing the tab-switching and context re-describing steps, so desktop work and study users who research and write in the same session would pick it for complex questions; public material offers no retention or payment evidence.

Entry and what to borrow

when AI-ifying an existing entry, copy its "Inspiration Explorer" split — don't make search a one-shot Q&A; make it lead the user to think further. Add follow-up directions, problem maps and related links beside the answer so users stay on the page instead of leaving after one reply. That turns the AI from an answering tool into a thinking guide.

Evidence and risk

Search ads first, subscriptions and the agent ecosystem as incremental revenue.; Per the Q3 2025 earnings call (Nov 18, 2025):; 12M subscribers to the Wenxin Assistant inside the Baidu app; Agent daily revenue for advertisers exceeded RMB 2… ① Where the 12M subscriber count goes — willingness to pay is the hardest proof metric; for AI search; ② Whether traffic holds through the "Wenxin Assistant" rebrand — the transition will; expose whether the product has its own mindshare; ③ Whether the RMB 25M/day agent revenue keeps growing — it de…

What this judgment rests on
Public fact

Turns Baidu search into a chat door: it answers, digs further, and keeps writing and image tools on the same page.

Workflow reasoning

Inference: versus opening links one by one and switching tools, it chains follow-up, reading and creation on one page, removing the tab-switching and context re-describing steps, so desktop work and study users who research and write in the same session would pick it for complex questions; public material offers no retention or payment evidence.

The unknown that could change the call

An English validation note will follow from the public evidence.

01 · Value Insufficient evidence

The product claims to help users complete: “Turns Baidu search into a chat door: it answers, digs further, and keeps writing and image tools on”. User evidence has not yet verified pain intensity or the cost of doing without it.

02 · Consensus Insufficient evidence

The assessment is recorded; an English explanation is pending.

03 · Model Insufficient evidence

The assessment is recorded; an English explanation is pending.

04 · Truth Insufficient evidence

The assessment is recorded; an English explanation is pending.

02

Chinese and English ecosystems

Market comparison

The Chinese–English market comparison is not complete yet. A conclusion follows only after its coverage and verifiable evidence are recorded.

03

60-second business read

The call and next move come first; the full read retains the evidence and counterevidence.

What it is in one line

Baidu Search's AI version: the "search box plus link list" becomes a conversation entry point that answers questions, offers "Inspiration Explorer" to think further, and squeezes AI reading, AI writing, AI slides, agents and AIGC creation into the same page.

Who built it

Baidu. The product started as "AI Search" on PC (chat.baidu.com) in January 2025, went through its "biggest revamp in a decade" in July 2025, and was rebranded "Baidu Wenxin Assistant" in October 2025, replacing the "Baidu AI Search" brand. It shares its foundation with ERNIE Bot (per Baidu Baike).

Read: this is Baidu wrapping its national-search-entry in an AI shell, not an independent startup. Its biggest asset is the daily search habit of hundreds of millions of Chinese users; its biggest burden is ad revenue — search companies are always the slowest to disrupt their own business.

What it actually does

  • AI search → conversational querying, multi-turn refinement, answer-first output with cited sources
  • Inspiration Explorer → after a question, branches out related angles and links to help you map the problem (marketed as unique)
  • AI reading → one-click summaries of links, images and oversized documents
  • AI writing / AI slides → 20+ writing styles; automatically generated well-structured slide decks from web content
  • Agents → an agent zone at the corner of the results page where you can @ domain-specific agents for vertical answers
  • AIGC creation → upgraded to eight modalities in October 2025: AI video, AI music, AI podcast, AI images and more; the search box becomes a creation entry
  • Digital human agents → open, real-time interactive search avatars launched October 2025

What it deliberately does not do: it does not spin off as a standalone product line. Every capability grows on the search box; open Baidu and it is there, no new app to download.

What old behavior it replaces

It replaces the twenty-year-old habit of "search equals a list of links." Type "2025 CNY brand cross-border marketing plan" into Baidu and you used to get dozens of URLs to open, compare and synthesize yourself. Now the same question returns a finished plan — target analysis, execution steps, blanks left for you to fill in.

It also replaces a second action: asking an AI meant opening ChatGPT or the ERNIE app separately — search and Q&A were two different entry points. Baidu merges them into one move. You are already on Baidu; you don't switch places. That is the "owner of the front door" playbook for AI.

It even touches the habit of rephrasing queries. With traditional search, a badly phrased question means retrying with new keywords. Inspiration Explorer instead offers directions to explore from your existing question, doing the "what am I actually asking" work for you.

Business model

Search ads first, subscriptions and the agent ecosystem as incremental revenue. Per the Q3 2025 earnings call (Nov 18, 2025):

  • 12M subscribers to the Wenxin Assistant inside the Baidu app
  • Agent daily revenue for advertisers exceeded RMB 25M
  • Conversation rounds up roughly 5x year-over-year

Read: the ad business means Baidu can never truly give the traffic away to AI answer pages — AI answers have no ad slots. Its balancing act is: free AI answers retain users, while subscriptions (12M) and agent ads (RMB 25M/day) regrow revenue on the AI side. That math only holds if AI answers don't cannibalize traditional ad clicks, and that contradiction is growing.

Hard numbers

  • traffic board: 44.85M monthly visits, +13.76% MoM (2026-08)
  • QuestMobile (Aug 2025 monthly): MAU 365M, No.1 in China's AI search
  • QuestMobile (Q3 2025): MAU 382M, No.1 for three consecutive quarters
  • Earnings call (2025-11-18): 12M subscribers, RMB 25M+ daily agent revenue, ~5x conversation rounds YoY
  • Rebranded "Baidu Wenxin Assistant" in October 2025

Four-way read

Dimension Call
Founder-product fit A company-level strategic product; fit depends on whether Baidu is willing to disrupt its own search business
Product insight Inspiration Explorer is a rare differentiator in AI search; turning the search box into a task entry is the right direction
Execution quality In-house accelerated model (5x faster generation than DeepSeek V3.1 at 70% cost) gives a real technical base
Timing The window is now, but Doubao, DeepSeek, Kimi and Yuanbao are all on the field; the race is for existing-user habits

The call

A case study in how an incumbent does AI — the traffic is real, and so is the anxiety.

Baidu AI Search's MAU (382M) and traffic board visits (44.85M) are both genuine, and QuestMobile has ranked it No.1 in AI search for three straight quarters. That No.1 rides on the installed base: hundreds of millions already use Baidu, and AI features keep a fraction of them. Compare that with Doubao's 3.8x doubling trajectory — the problem is not that nobody uses Baidu AI Search, it is where new users come from. Its users arrive through the old "Baidu" entry, not through a new "AI search" mindshare.

The rebrand to "Wenxin Assistant" is a signal: Baidu is trying to promote the product name from a feature ("AI search") to a product ("assistant"), because it knows the search box alone is no longer a strong enough entry. 12M subscribers is top-tier among domestic LLM products and proves people will pay for AI answers.

The +13.76% MoM is healthy, no spikes and no shrinkage — but in this competitive field, defending the No.1 slot in existing traffic is not the same as winning increment.

What to watch next

① Where the 12M subscriber count goes — willingness to pay is the hardest proof metric for AI search ② Whether traffic holds through the "Wenxin Assistant" rebrand — the transition will expose whether the product has its own mindshare ③ Whether the RMB 25M/day agent revenue keeps growing — it decides whether the second revenue curve is real

What you can take from it

Product logic: when AI-ifying an existing entry, copy its "Inspiration Explorer" split — don't make search a one-shot Q&A; make it lead the user to think further. Add follow-up directions, problem maps and related links beside the answer so users stay on the page instead of leaving after one reply. That turns the AI from an answering tool into a thinking guide.

Product logic (entry-point merge): if you already own an entry users visit anyway (installed base, legacy traffic, default page), the right way to add AI is to grow the capability on the entry — not to make users download a separate app. Baidu's "no need to switch places" move applies to any product with existing reach.

Verdict

Worth watching. The traffic and data are real and the top of the market is set, but the incremental story (subscriptions plus agent revenue) is not fully told. It is the best live sample of how an incumbent does AI — real entry advantage and a real business-model contradiction.

05

Go from the product name to primary material

Use these searches when the official site is missing or the current link is only a lead.